--TYecfFk8j8mZq+dy
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, Jul 25, 2005 at 09:19:42PM +0200, Martin Husemann wrote:
> On Mon, Jul 25, 2005 at 08:22:39PM +0200, Zeljko Vrba wrote:
> > quite often everything just stops due to ioflush. And I mean, STOPS.
>=20
> What are you doing? NFS over a 9600 bps SLIP link?

No, he's not.

The problem is that our UBC system doesn't put back-pressure on programs.=
=20
You can dirty pages with mmap very quickly yet not trigger back-writing of=
=20
the pages. So you fill up all of RAM, then have to wait for the ioflush=20
system to turn on and flush data out.

Part of the problem is that we flush "files". Thus we will flush or not=20
flush a file on a file by file basis. This works great if we have a lot of=
=20
writes to different files; we will hit on different files, and keep a=20
steady stream of data going to the disks.

The problem is if all or most all the dirtying is happening in one file.=20
We do nothing until we try to clean that one file, then we clean=20
everything at once. And we drag the system to its knees in the mean time.

Take care,

Bill

--TYecfFk8j8mZq+dy
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (NetBSD)

iD8DBQFC5T2mWz+3JHUci9cRAu/VAJ9Wng/w0GuCiB0+qZvpGkM8TVExNgCeMl6k
5KtmIP1M5cxBhlpV7q3uuZo=
=fuBI
-----END PGP SIGNATURE-----

--TYecfFk8j8mZq+dy--